tokencalc

CLI tool and Python library for calculating Claude API token costs. Track your Claude Code spending, compare models, plan budgets, and export to CSV.

Install

pip install tokencalc

Claude Code usage tracking

See how much your Claude Code sessions are costing you:

# latest session breakdown (tokens, cache, cost)
tokencalc cc

# list recent sessions with costs
tokencalc cc list

# all-time usage totals
tokencalc cc all

# last 5 sessions in detail
tokencalc cc last 5

  Claude Opus 4  2026-03-13 20:21
  156 API calls

  Input:                 188 tokens  ->  $0.002820
  Output:              32.2k tokens  ->  $2.41
  Cache write:        115.0k tokens  ->  $2.16
  Cache read:          7.44M tokens  ->  $11.16
  --------------------------------------------------
  Total:               7.58M tokens  ->  $15.73

  Cache saved you $100.40 (vs no caching)

Reads directly from ~/.claude/projects/ session logs. No setup needed -- just have Claude Code installed.

Quick calculations

tokencalc 50000 10000

  Claude Sonnet 4
  Input:     50.0k  tokens  ->  $0.1500
  Output:    10.0k  tokens  ->  $0.1500
  Total:     60.0k  tokens  ->  $0.3000

Specify a model:

tokencalc 50k 10k opus-4

Compare across all models:

tokencalc --compare 100k 20k

Interactive mode

tokencalc

Drops into a REPL with all commands available:

Token counts accept shorthand: 10k, 1.5M, 10,000.

Use as a library

from tokencalc import calc_cost, estimate_tokens, MODELS, parse_session, find_session_files

# calculate cost with cache breakdown
cost = calc_cost(
    input_tokens=50_000,
    output_tokens=10_000,
    model="sonnet-4",
    cache_write_tokens=5_000,
    cache_read_tokens=100_000,
)
print(f"Total: ${cost['total']:.4f}")

# parse a Claude Code session
sessions = find_session_files()
data = parse_session(sessions[0]["path"])
print(f"Session cost: ${data['cost']['total']:.2f}")

# estimate tokens from text
tokens = estimate_tokens("some long prompt text here...")

# list models
for key, info in MODELS.items():
    print(f"{info['name']}: ${info['input']}/M in, ${info['output']}/M out")

Supported models

Pricing from the Anthropic docs. Open an issue or PR if anything's out of date.

License

MIT